Re-examination of Bagnold's empirical bedload formulae